NVIDIA A100

NVIDIA Graphics Card

The NVIDIA A100 Tensor Core GPU is the data centre compute workhorse of the Ampere generation, with 6,912 CUDA cores and 432 third-generation Tensor Cores paired with HBM2e high-bandwidth memory (PCIe models offer 40GB or 80GB) reaching up to 1,935 GB/s. Multi-Instance GPU (MIG) can partition one card into up to seven independent instances so a single GPU serves several workloads at once, while an NVLink bridge provides 600 GB/s of card-to-card bandwidth. The passively cooled dual-slot form factor supports high-density rack server deployment, making it a proven choice for deep learning training and high performance computing.

Specifications

ArchitectureNVIDIA Ampere architecture (GA100)
3rd-generation Tensor Cores
Compute Cores6,912 NVIDIA CUDA Cores
432 3rd-generation Tensor Cores
GPU MemoryHBM2e high-bandwidth memory
40GB or 80GB on PCIe models, up to 1,935 GB/s
System InterfacePCIe 4.0 x16
InterconnectNVLink bridge support, 600 GB/s card-to-card bandwidth
VirtualisationMulti-Instance GPU (MIG), up to 7 independent instances
NVIDIA vGPU support
Power250W (40GB) / 300W (80GB) max power consumption
Cooling & Form FactorPassive cooling (chassis airflow)
Dual-slot full-height full-length, for rack servers
Software SupportCUDA, cuDNN, NCCL, TensorRT, NVIDIA AI Enterprise
Use CasesDeep learning training, large model inference, HPC, scientific simulation, data analytics
載入中…